Physical exercise may reduce buildup of Atg9 vesicles in motor neurons depending on age in disease models

All mice used in this study are listed in Table S1. Mice were handled according to the institutional guidelines of the University Clinic Wuerzburg and the German federal law of animal protection. Mice were kept at constant room temperature and humidity at a 12/12-h light condition with food and water offered ad libitum. Mice aged 3 and 12 months were split into sedentary and running groups for voluntary physical exercise. Exercising mice were individually placed within a new cage containing a running wheel (1800/50, Ugo Basile, Germonio, Italy) and left undisturbed to perform voluntary physical exercise for either four hours or four weeks consecutively. Mice from the sedentary group were individualized as well and housed without running wheels. Physical exercise was evaluated based on a daily recording of running wheel revolutions. Mice which performed less than 800 revolutions within four hours were excluded from the experiments. Mice which performed less than 800 revolutions daily on three consecutive days were excluded from the experiments. All mice had access to water and food ad libitum during the entire time. Grip-strength was measured for all mice before and after four weeks of physical exercise. Measurements were carried out on two consecutive days. Grip strength (93153, Ametek, Berlin, Germany) was measured at least 3 times per mouse.

Mice were sacrificed, tibialis anterior (TA) and gastrocnemius (GAS) muscles removed, fibers separated and fixed for two hours in 4% paraformaldehyde (PFA). Fibers were washed three times with PBS, and once with 0.1 mol/L glycine for 15 min. Samples were incubated in 10% donkey serum with 0.3% Triton X-100 in TBS-T for two hours at room temperature (RT). Three subsequent washing steps with TBS-T containing 0.1% Triton X-100 were performed and samples were incubated with primary antibodies in blocking solution for one day at 4 °C. Samples were washed three times, incubated with secondary antibodies at RT for 2 h, and washed with PBS. Fibers were carefully separated and evenly distributed on an object slide before mounting with FlourSave (Merck, 345789, Darmstadt, Germany). To visualize postsynaptic acetylcholine receptors (AChRs), α-bungarotoxin (BTX) conjugated to Alexa-488 from Invitrogen (B13422, Dreieich, Germany) was used. Secondary antibodies were obtained from Jackson Immuno-Research Laboratories. Primary antibodies are listed in Table S2. Secondary antibodies are listed in Table S3.

Mice were euthanized for 5 min with CO₂ and trans-cardially perfused with 4% PFA. Spinal cords were removed after perfusion, post-fixed for two hours with 4% PFA, and sections cut at a Leica VT1000S Vibratome with 40-µm thickness. Free-floating sections were washed three times with PBS, once with 0.1 mol/L glycine for 15 min and once with ammonium-acetate for 30 min. Subsequently, samples were blocked in 10% donkey serum with 0.3% Triton X-100 in TBS-T for two hours at RT and incubated with primary antibodies for two days at 4 °C. Samples were washed three times, incubated with secondary antibodies at RT for two hours, and washed thoroughly with PBS. Afterwards, samples were mounted with FlourSave (Merck, 345789). Secondary antibodies were obtained from Jackson Immuno-Research Laboratories. DAPI (Sigma-Aldrich, D9542-5MG, Darmstadt, Germany) was used for nucleic acid staining. Primary antibodies are listed in Table S2. Secondary antibodies are listed in Table S3.

Images of NMJs were randomized and individually analyzed for shape, integrity and balloon-like accumulations at the presynapses. Only presynaptic staining colocalized with postsynaptic marker BTX was analyzed. NMJ shape was categorized as unaffected, fragmented with partial loss of presynaptic staining compared to corresponding postsynaptic staining, and denervated with more than 90% of presynaptic marker gone. Balloon-like structures were counted as presynaptic accumulation with an area of more than 3 µm diameter.

All spinal cord images were randomized and analyzed in an unbiased way. The Deepflash 0.2.3 learning algorithm was used to identify fluorescence signals [21]. Spinal cord sections were observed under a confocal scanning microscope (Olympus FV1000) and raw images were captured and then converted to 8 bits and scaled to 1024 × 1024 pixels resolution by ImageJ. 20 × magnification images were stitched with the ImageJ stitching plugin. Spinal cord samples with low abundance of targeted signal were imaged with 40 × magnification. For evaluation of the Deepflash data, five image training data sets were created manually for model generation by experts. Ground truth estimation was performed with masks from five experts. Different models were created for respective experiments and fluorescence patterns. Model performance metrics were based on the inbuild dice score for semantic segmentation, with retraining of models until a dice score of at least 0.7 was reached. Furthermore, resulting masks from model training were cross-checked by experts to ensure model quality. Resulting images were frequently controlled by random sampling and sporadic false positive values resulting from cell body detection were manually removed. Spinal cord gray matter and NMJ areas were measured using free-hand sections in ImageJ. Spinal cord sections were counted from one side including the ventral horn up to the central canal. Generated image masks were analyzed with the particle analyze function, ignoring particles with a size below 10 pixel and outside the gray matter area. Measured particles were normalized to the respective gray matter area and resolution and displayed as number per area. Graphic creation and statistics were performed with OriginPro 2021b Academic. Outliers were removed upon statistical significance by Grubbs' outlier test.

Images were processed as described above. MNs and NMJs were manually traced for area calculation. Corresponding masks from GFP and RFP were colocalized, and double-positive signals were counted as autophagosomes. Autophagosome numbers were subtracted from the total RFP⁺ signals and the remaining RFP⁺ signals were counted as autolysosomes. Signals were normalized to respective areas. Outliers were removed upon statistical significance by Grubbs' outlier test.

Whole spinal cord tissues were dissected and immediately frozen in liquid nitrogen. Fractionation was performed following the protocol from Wirths [22]. Briefly, tissue was homogenized in homogenization buffer and centrifuged for 10 min at 1000 g. The buffer contained 0.32 mol/L sucrose, 5 mmol/L HEPES, protease inhibitor (5892970001, Millipore Sigma, Darmstadt, Germany) and phosphatase inhibitor (4906837001, Millipore Sigma). The supernatant was collected (Input) and centrifuged at 17,000 g for 30 min. The pellet was washed with PBS and centrifuged at 17,000 g for 30 min. The supernatant was transferred into polycarbonate tubes (252240, Beranek, Nußloch, Germany) and centrifuged at 100,000 g for 1 h. The supernatant was collected (Cytosol), and the pellet was washed with PBS. Finally, the washed pellet from the 100,000 g centrifugation (100 k) was resuspended in PBS. The input from 3–4 spinal cord homogenates was pooled to obtain sufficient material for the 100 k pellet. Protein yield was determined by the Bradford assay. All centrifugation steps were carried out at 4 °C.

Equal amounts of protein were separated by SDS-PAGE, and transferred to PVDF membranes (1620177, Bio-Rad, Dreieich, Germany) (120 V, 45 min, 4 °C). Membranes were blocked in TBS-T with 5% milk powder for 2 h at RT, probed with primary antibodies overnight at 4 °C, and incubated with horseradish peroxidase-conjugated secondary antibodies for 1 h at RT. Membranes were washed three times in TBS-T for 15 min and incubated for 5 min with developer reagents (Immobilon, WBKLS0500, Darmstadt, Germany). Primary antibodies used for Western blot analysis are listed in Table S4. Peroxidase-conjugated secondary antibodies against mouse (715–005-150) and rabbit (711–005-152) were obtained from Jackson Immuno-Research Laboratories (Cambridgeshire, UK). Peroxidase-conjugated secondary antibody against goat (AP180P) was obtained from Merck.

For ultrastructure analysis, mice were transcardially perfused following a modified Forssman perfusion protocol [23]. Two fixation solutions were subsequently used, the first containing 1.5% PFA with 1.5% glutaraldehyde and the second containing 3% PFA, 3% glutaraldehyde and 0.05% picric acid in water. Mice were slowly perfused with 75 mL fixative one and 100 mL fixative two. Afterward, tissue was dissected and post-fixed in 4% PFA with 4% glutaraldehyde for two hours at 4 °C. Samples were washed with 0.1 mol/L phosphate buffer and contrasted with 2% osmium tetroxide for 2.5 h at 4 °C. The staining and embedding protocol was modified after Mulisch and Welsch [24]. Briefly, fixed samples were washed in 0.1 mol/L phosphate buffer, incubated in 0.5% aqueous uranyl acetate overnight, washed and dehydrated in serial ethanol concentrations from 50% to 100%. Dehydrated samples were placed two times in propylene oxide for 30 min each at RT, followed by a 1:1 mix of propylene oxide and epoxy resin overnight. The samples were then embedded in epoxy resin. Cured samples were sliced to 70-nm sections and imaged under a transmission electron microscope (JEOL JEM-2100, Freising, Germany) at 200 kV with a TVIPS F416 digital camera.

Mice were sacrificed, spinal cord tissue removed, and immediately frozen in tissue tack (Sakura Finetek, 4583, Umkirch, Germany) at − 80 °C. Samples were cut with a cryostat (Leica CM 1950) at 5 µm thickness and mounted on coverslips coated with poly-L-lysine. The sections were fixed with 4% PFA for 15 min at RT, followed by incubation with 0.1 mol/L glycine for 15 min. Samples were blocked in 10% donkey serum with 0.3% Triton X-100 in TBS-T for two hours at RT and incubated with primary antibodies for one day at 4 °C. After rinsing with PBS, secondary antibodies were applied for two hours, followed by three washing steps with PBS. The dual-color dSTORM images were acquired using a Zeiss Observer Z.1 inverted light microscope (Carl Zeiss AG) at the Rudolf Virchow Center for Integrative and Translational Bioimaging (Würzburg, Germany). The design was adapted for single-molecule localization microscopy (SMLM) as shown in Table S5. The blinking buffer contained 125 mmol/L cysteamine hydrochloride (Sigma, M6500), 20 mmol/L D-glucose (Sigma, G7528), 0.55 mg/mL glucose oxidase (Roth, 60281), and 0.011 mg/mL catalase (Sigma, C1345) in PBS, with pH adjusted to 7.7 using a 5 mol/L KOH solution [25]. The order of acquisition was important to avoid unnecessary bleaching. First, A647 was acquired, followed by Cy3. The frames of the dSTORM images were reconstructed using the ThunderSTORM Plugin for ImageJ [26, 27], generating super-resolved images with a pixel size of 102 nm. Detailed camera setups and image processing specifications are listed in Table S6.

Spinal MNs from embryonic mice were isolated and cultured as previously described [28, 29]. Briefly, spinal cords from E13 embryos were dissected and incubated for 15 min in 0.05% trypsin in Hank's balanced salt solution. Cells were triturated and incubated in Neurobasal medium, supplemented with 100 µg/mL penicillin–streptomycin-glutamine (Gibco, 10378016) on Nunclon plates (Thermo Fisher Scientific, 150350, Dreieich, Germany) pre-coated with antibodies against the p75 NGF receptor (MLR2, a kind gift from Robert Rush, Flinders University, Adelaide, Australia) for 45 min. After washing with Neurobasal medium, the remaining MNs were recovered with a depolarization solution (0.8% NaCl, 35 mmol/L KCl and 2 mmol/L CaCl₂). Cells were subsequently collected in the MN medium (2% horse serum, 1 × B-27 in Neurobasal medium with Glutamax). Before plating, MNs were transduced with lentiviral vectors for expression of ChR2-YFP and RFP-LC3. Cells were plated on four-well dishes (Greiner Bio-One, 627170, Schwerte, Germany) pre-coated with poly-ornithine/laminin (Sigma-Aldrich, L2020-1MG). MNs were cultured with the neurotrophic factor BDNF (10 ng/mL) following a medium change every second day.

Stimulation experiments were performed on day 7. Control samples were protected from light at least one hour before fixation. Stimulation samples were placed beneath a blue light source (470 nm, 10% laser intensity) and exposed for 2 min with 0.2 Hz followed by a recovery time of 8 min in an incubator. Exposure was repeated three times and MNs were subsequently fixed in 4% PFA for 15 min.

Fixed MNs were washed once with TBS-T and blocked with 10% horse serum in TBS-T for one hour. The samples were incubated in primary antibody solution containing TBS-T overnight at 4 °C. Afterwards, samples were washed three times with PBS and incubated with secondary antibody solution in PBS for one hour. MNs were washed three times with PBS and mounted in FlourSave. Alexa-488-, Cy3- and Cy5-conjugated secondary antibodies were obtained from Jackson Immuno-Research Laboratories. Primary antibodies are listed in Table S2.

Based on the activity-coupled cycling of Atg9⁺ vesicles [10], we investigated whether exercise-induced neuronal activity could change the accumulation of Atg9⁺ vesicles in the spinal cord. Therefore, we analyzed the Atg9⁺ accumulations in the gray matter of spinal cord sections upon four hours of voluntary physical exercise using a deep learning segmentation pipeline [21]. After four hours of voluntary exercise, we observed only minor, non-significant differences in the size and number of Atg9⁺ clusters (Fig. S1), suggesting that four hours are not sufficient to modulate the accumulations.

Despite the inability to clear Atg9⁺ clusters with progressive age, we wondered whether Atg9⁺ clusters reached a terminal, dysfunctional stage over time. Therefore, we analyzed the composition of Atg9⁺ clusters using immunohistochemistry. Based on previous work [11], we stained for the endo/lysosomal marker Lamp1 and the SV marker Synaptophysin. The majority of Atg9⁺ clusters stained positive for the late endosomal/lysosomal marker Lamp1, while only a minor subset was positive for the SV marker Synaptophysin (Fig. 4b, c). This finding aligns with the previously reported heterogeneity between SV and Atg9⁺ vesicle populations [11]. At the ultrastructural level, the vesicle clusters appeared as accumulations of individual, densely packed vesicles (Fig. S3a). Next, we further characterized the Atg9⁺ clusters, which overlapped with the late endosomal/lysosomal marker Lamp1. To address whether the Atg9⁺ Lamp1⁺ clusters represent a homogenous population with individual vesicles carrying both Atg9 and Lamp1 or a heterogenous population with individual vesicles carrying either Atg9 or Lamp1, we performed super-resolution microscopy (dSTORM). Using dSTORM, we detected individual ring-like structures, which carried both Atg9 and Lamp1 with a similar size range as the vesicles we detected by electron microscopy (Fig. S3b).

To determine whether age influences the Atg9⁺ cluster composition, we compared the Atg9⁺ clusters of 3- and 12-month-old mice with versus without physical exercise. While we did not observe any difference within 3-month-old mice (Fig. 4d, e), the proportion of the Atg9⁺ clusters positive for Synaptophysin significantly increased after exercise in 12-month-old mice (Fig. 4f, g). These findings suggest that physical exercise increases the sorting of SVs into existing Atg9⁺ clusters. While the overall reduction of Atg9⁺ clusters with exercise prevented the detection of vesicle composition changes in 3-month-old mice, sorting of SVs into the Atg9⁺ clusters is still ongoing with age, although its removal is no longer possible.

Neuronal activity requires the cycling of SVs and the subsequent removal of damaged vesicles through the autophagy machinery [6]. Based on the enhanced cycling of SVs during elevated neuronal activity [7], we investigated if and how physical exercise induces MN autophagy in young and aged mice. Due to the previously described age-dependent decline of autophagy biogenesis [33], we hypothesized that the failure to remove the Atg9-containing vesicles might result from impaired autophagy in aged mice.

To analyze changes in the autophagy levels, we crossed Plekhg5-deficient mice with mRFP-GFP-LC3 (Light chain 3) expressing mice. Fusion between autophagosomes and lysosomes leads to the acidification of the autolysosomes, which quenches the GFP signal [34]. Therefore, the tandem reporter enables differentiation between autophagosomes (RFP⁺GFP⁺) and autolysosomes (RFP⁺GFP⁻). This strategy allows quantification of the autophagic flux, which is particularly important due to the rapid turnover of autophagosomes in neurons [35]. First, we validated the expression of LC3-RFP-GFP in NMJs and MN somata. In MN somata, we mostly detected RFP⁺GFP⁻ vesicles, with only very few RFP⁺GFP⁺ vesicles (Fig. S4a). In contrast, comparable numbers of RFP⁺GFP⁺ and RFP⁺GFP⁻ vesicles were detectable within axon terminals at NMJs (Fig. S4b). These data align with the previously reported spatial separation of the autophagic process in neurons. Autophagosome biogenesis mostly occurs in distal axons, followed by maturation during retrograde transport and fusion with lysosomes at the somata [16, 36, 37]. We confirmed the presence of autophagosomes and the absence of lysosomes at MN axon terminals by electron microscopy. Vice versa, autophagosomes were hardly detectable in MN somata, while lysosomes were readily detectable (Fig. S4c).

To confirm that autophagy induction depends on neuronal activity in a cell-autonomous manner, we expressed ChR2-YFP and RFP-LC3 in cultured primary mouse MNs. We analyzed the number of autophagosomes within axons after light stimulation and found an increased number of both SVs and autophagosomes (Fig. S4d–g). Furthermore, the number of autophagosomes carrying SVs as cargo also increased with blue light stimulation (Fig. S4g).

Next, we asked whether physical exercise also induced autophagy in 3-month-old Plekhg5-deficient mice. Therefore, we analyzed the number of autophagosomes and autolysosomes in MN somata from sedentary and exercising mice (Fig. 5d, e). In line with previous work [16], we detected a lower number of autolysosomes in sedentary Plekhg5-deficient mice compared to sedentary wildtype mice. Similar to wild-type mice, we found no change in the number of autophagosomes but an increased number of autolysosomes after physical exercise, strongly indicating that neuronal activity provides a mechanism to bypass the autophagy defects in Plekhg5-deficient mice. To corroborate our results biochemically, we examined the endogenous level of LC3 by Western blot analysis of spinal cord lysates (Fig. 5f, g). In the 3-month-old wild-type and Plekhg5-deficient mice, we detected a modest but robust decrease in LC3-II upon physical exercise. Notably, this decrease in LC3-II upon physical exercise was not detectable in 12-month-old mice.

The age-dependent autophagy decline in neurons has been linked to WIPI2B (WD repeat domain, phosphoinositide interacting 2B) [33]. Ectopically expressed WIPI2B restores autophagosome biogenesis in aged neurons [33]. This rescue requires the dynamic phosphorylation of WIPI2. Therefore, we performed membrane fractionation of spinal cords from 3-month-old wild-type and Plekhg5-deficient mice and analyzed the levels of WIPI2 and its phosphorylated form by Western blot (Fig. 5h–j). After homogenization and differential centrifugation of the tissue, we detected a clear enrichment of membrane proteins in the 100,000 g centrifugation pellet (membrane pellet), confirming the successful separation of the cytosolic and membrane fractions (Fig. 5h). Within the membrane fraction obtained from Plekhg5-deficient mice, we found an increased amount of phosphorylated-WIPI2 (p-WIPI2) compared to the wildtype (Fig. 5i). Interestingly, the total WIPI2 showed a comparable increase, suggesting an accumulation of WIPI2B in its phosphorylated form at isolation membranes. To assess whether the WIPI2 levels changed upon physical exercise, we analyzed the membrane fractions of sedentary and running mice at 3 and 12 months (Fig. 5j). Whereas no striking differences in the membrane levels of WIPI2 were detectable in the wildtype control samples, we observed a marked reduction of p-WIPI2 and WIPI2 levels in 3-month-old Plekhg5-deficient mice upon physical exercise (Fig. 5j). In 12-month-old animals, no decrease in the WIPI2 accumulation was apparent upon physical exercise. In the cytosolic fractions, we did not observe any striking differences. To further corroborate our findings, we also analyzed the levels of Atg9 in the membrane fraction. In line with our immunohistochemical data, we found a decrease of Atg9 in 3-month-old but not 12-month-old Plekhg5-deficient mice upon physical exercise (Fig. S4h).

In summary, these data indicate that short-term physical exercise accelerates the autophagic flux in MNs of 3-month-old mice, reflected by increased autolysosomes at the MN somata and reduced LC3-II in spinal cord lysates. Interestingly, physical exercise also triggered autophagy in Plekhg5-deficient mice, strongly indicating that Plekhg5 mediates SV turnover under basal conditions, uncoupled from neuronal activity, leading to the removal of Atg9⁺ clusters. Furthermore, our data point to WIPI2B as a major target to bypass the Plekhg5 deficiency-related autophagy impairment by physical exercise. While 3-month-old Plekhg5-deficient mice showed reduced WIPI2 levels following exercise, the effect was not apparent in 12-month-old mice, suggesting that WIPI2 accumulation upon Plekhg5 depletion can be reversed by neuronal activity.

To examine whether physical exercise also leads to a reduction in the number of Atg9⁺ vesicles in SOD1^G93A mice, we subjected SOD1^G93A mice to four weeks of voluntary exercise and analyzed the number of Atg9⁺ clusters in the spinal cord sections (Fig. 6d, e). Compared to sedentary SOD1^G93A mice, we detected a marked reduction of Atg9⁺ clusters in all mice performing physical exercise (Fig. 6e). To follow up on our previous observation of increased sorting of SVs into the Atg9⁺ compartment, we analyzed the co-localization of SVs with Atg9⁺ clusters. Interestingly, SVs were rarely detectable in both sedentary and exercising mice (Fig. 6f, g). Furthermore, most Atg9⁺ clusters stained negative for the Lamp1 marker. Taken together, we conclude that the clustering of Atg9⁺ vesicles represents a broader disease mechanism in MND, which depends on neuronal activity. Furthermore, clearing of Atg9⁺ clusters by physical exercise is a valid approach to reducing the intracellular burden of accumulating vesicles and improving the histopathological disease progression.